From Patterns to Adaptive Representations
From recognizing patterns to constructing the right representation for a problem. Five lenses on how AI can learn, generalize and adapt.
Read the essayImagine · Decision lab
Two Gates
JEPA × Diffusion × Active Inference
Explore when information is worth a detour. Compare imagined futures, uncertainty and action in a small decision environment.
Try this
A question for the model
Compare a direct route with an information-seeking route. Inspect what the agent knows before and after observing.
Read the model
What you are exploring
A toy decision environment with a JEPA-style latent model and an operational approximation of expected free energy. It is a teaching experiment, not a full implementation of Active Inference.
Continue the loop
Connected ideas
AI Alignment Is an Architecture Problem
Why alignment may depend on how intelligence is organized, and how it learns what matters under pressure.
Read the essay
Six Human Needs
Agents explore a landscape using different mixtures of certainty, variety, connection, significance, growth and contribution.
Explore the experiment